DIM3 CUDA SAMPLE CODE DRIVERS

Cuda is an extension of the C programming language that is created and maintained by Nvidia. Using GPU architecture as a guide, we introduce GPU programming using Cuda. We begin with a brief introduction to GPU architecture. In this document, we introduce GPU-style parallelism using Cuda and explore some solutions to the 3sum problem utilizing a GPU co-processor. One may work through the document alone, or instructors can prepare lectures based on the information and samples.

DIM3 CUDA SAMPLE CODE DRIVERS

Cuda drivers and the Cuda development kit (nvcc) installed on that computer.Access to a computer with a Cuda enabled GPU.Ability to write, compile, and execute a C/C++ program.Describe the memory hierarchy of the GPU.

Describe the concepts of grid, grid dimensions, blocks, and threads and how they relate.

Explain the aspects of thread communication that are unique to the GPU.

Describe the basic outline of a Cuda program.

Explain the concept of thread divergence.

Design algorithms with threads that take advantage of GPU architectures.

Identify portions of an algorithm that are good candidates for parallelization.

Define the concept of a thread block in Cuda.

Explain the major, conceptual, architectural differences between a CPU and a GPU processor.

Describe the general architecture of a GPU.

Nvcc -arch=sm_11 SOURCE.cu -o EXENAME (for architecture version 1.1)Ĭontent Title: Introduction to the MapReduce Model of Parallelism Nvcc -arch=sm_20 SOURCE.cu -o EXENAME (for architecture version 2.0) Thus, to compile these programs, the command will look like: Technically, atomic operations using integer arguments was first available in leve 1.1. In general, architecture version 2.0 introduced a lot of very useful functionalility, and has been around sufficiently long that one can usually assume their card supports this level of functionality. The Cuda code in these examples that use the atomicAdd() function must be compiled with compiler flags that support atomic functions.

Dim3 cuda sample code